[sldev] SSE2 on MSVC

Erik Anderson odysseus654 at gmail.com
Sun Jun 17 18:23:56 PDT 2007


According to the docs, "By default, SSE2 implementation (using
_set_SSE2_enable) is enabled on processors that support it".  Does this
change cause an improvement in performance at all?

At this point I believe that SSE2 is a number of CPU instructions designed
to "help speed 3D graphical calculations" or somesuch (or was that MMX?).
If a program is compiled to use SSE2, then it (can) run faster on the
SSE2-enabled processor than the original version would and cause spectacular
crashes on the non-SSE2-enabled processor.  Thus our discussion here about
whether or not to support every CPU model.

Does anyone have a good idea as to where these SSE2 commands would improve
SL the most?  Considering that the CRT appears to have multiple
implementations of the math libraries, one that uses SSE2 and the other that
doesn't, would it be helpful to have multiple implementations of some of our
functions?

On 6/17/07, Nicholaz Beresford <nicholaz at blueflash.cc> wrote:
>
>
> Dunno if the code bit below makes any sense (I haven't got
> much idea what SSE2 is, despite having read it on Wikipedia),
> but what do you think?
>
>         //
>         // enable MSFT SSE2 extensions
>         //
>         CProcessor cproc;
>         const ProcessorInfo *pinfo= cproc.GetCPUInfo();
>         int sse2= FALSE;
>         if (pinfo->_Ext.SSE2_StreamingSIMD2_Extensions) {
>                 //
> http://msdn2.microsoft.com/en-us/library/bthd138d(VS.80).aspx
>                 sse2= _set_SSE2_enable(TRUE);
>         }
>         llinfos << "running with" << (!sse2 ? "out" : "" ) << "
> _set_sse2_enabled " << llendl;
>
>
>
> Nick
>
>
> --
> Second Life from the inside out:
> http://nicholaz-beresford.blogspot.com/
> _______________________________________________
> Click here to unsubscribe or manage your list subscription:
> /index.html
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.secondlife.com/pipermail/sldev/attachments/20070617/9ada6f20/attachment.htm


More information about the SLDev mailing list